January 17, 2018
Machine Learning vs Programing
Machine Learning vs Statistics
linear \(\Rightarrow\) non-linear
additive \(\Rightarrow\) interactions
theory-driven \(\Rightarrow\) optimization-driven
Black Box Problem
A startup wants you to predict wine quality from its chemical composition.
Let’s predict wine quality
Step 1: Find data
Free online dataset about red and white variants of the Portuguese “Vinho Verde” wine.
Target: Quality from 1 to 10 Features: Acidity, alcohol …
P. Cortez, A. Cerdeira, F. Almeida, T. Matos and J. Reis. Modeling wine preferences by data mining from physicochemical properties. In Decision Support Systems, Elsevier, 47(4):547-553, 2009.
Step 1: Find data
TODO: Draw image with features.
1 - fixed acidity 2 - volatile acidity 3 - citric acid 4 - residual sugar 5 - chlorides 6 - free sulfur dioxide 7 - total sulfur dioxide 8 - density 9 - pH 10 - sulphates 11 - alcohol Output variable (based on sensory data): 12 - quality (score between 0 and 10)
Draw: Volatile acidity as polish remover, pH as pH paper, alchohol as liqour bottle, chlorides as salt prinkler, citric acid as lemon, sulfur as mushroom?,
Step 1: Find data
Step 2: Throw ML on your data
10x CV Test linear regression model, decision tree, random forest
Step 2: Throw ML on your data
| learner.id | mae.test.mean |
|---|---|
| regr.ranger | 0.4353923 |
| regr.lm | 0.5696394 |
| regr.rpart | 0.6015873 |
=> The random forest (ranger) is the best model.
Step 2: Throw ML on your data
Step 3: Profit
Client: “We would love to learn some insights.”
Looking inside the black box
What are the most important features?
How do features affect predictions?
How do features affect predictions?
Interactions with type of wine?
Interactions with type of wine?
Rule of thumb for wine quality?
Exceptionally bad wine
A customer gets a really bad prediction. What was the reason?
| 5589 | |
|---|---|
| type | red |
| fixed.acidity | 7.4 |
| volatile.acidity | 1.185 |
| citric.acid | 0 |
| residual.sugar | 4.25 |
| chlorides | 0.097 |
| free.sulfur.dioxide | 5 |
| total.sulfur.dioxide | 14 |
| density | 0.9966 |
| pH | 3.63 |
| sulphates | 0.54 |
| alcohol | 10.7 |
| quality | 3 |
## [1] 3.7628
Shapley Value
The best wine
| 877 | |
|---|---|
| type | white |
| fixed.acidity | 6.9 |
| volatile.acidity | 0.36 |
| citric.acid | 0.34 |
| residual.sugar | 4.2 |
| chlorides | 0.018 |
| free.sulfur.dioxide | 57 |
| total.sulfur.dioxide | 119 |
| density | 0.9898 |
| pH | 3.28 |
| sulphates | 0.36 |
| alcohol | 12.7 |
| quality | 9 |
## [1] 8.124267
The best wine
Technique: Shapley Values
What tools do we have?
Interpretable Models
Interpretable Models
Intepretable Model: Linear Regression
Intepretable Model: Decision Tree
Interpretable Model: Decision Rules
IF \(90m^2\leq \text{size} < 110m^2\) AND location \(=\) “good” THEN rent is between 1540 and 1890 EUR
Model-specific methods
Model-specific methods
Model-specific methods
TODO: Example for CNNs
Model-specific methods
TODO: Example for text (RNNs and attention?)
Model-agnostic methods
Model-agnostic methods
Model-agnostic Methods
TODO: PDP gif
Model-agnostic methods
TODO: Feature importance figure
Model-agnostic methods: Global Surrogate
Model-agnostic methods: Local Surrogate
Example-based Methods
Example-focused Methods
Example-focused Methods
TODO: Graphic for counterfactuals
Example-focused Methods
TODO: Graphic for prototypes
Interested in learning more?
More on interpretable machine learning in my book http://christophm.github.io/interpretable-ml-book/.
